Wind Resource Assessment Storm Hills, NWT Wind Resource Assessment Storm Hills, NWT Report prepared...

111
1 Wind Resource Assessment Storm Hills, NWT Report prepared by Adrian Matangi Wind Resource Assessment - Sea Breeze Power Corp. May 9, 2014

Transcript of Wind Resource Assessment Storm Hills, NWT Wind Resource Assessment Storm Hills, NWT Report prepared...

1

Wind Resource Assessment

Storm Hills, NWT

Report prepared by Adrian Matangi

Wind Resource Assessment - Sea Breeze Power Corp.

May 9, 2014

2

1 Contents 2 Executive Summary ............................................................................................................................................... 4

2.1 Analysis Conclusions .................................................................................................................................... 7

2.2 Recommendations and Next Steps ............................................................................................................ 11

3 Site Description ................................................................................................................................................... 13

3.1 General Regional Geography and Climatology .......................................................................................... 13

3.2 The Data Set ............................................................................................................................................... 20

3.3 Sensor Models and the Data Format ......................................................................................................... 20

4 Quality Control (QC) ............................................................................................................................................ 22

4.1 Data Flagging and Inspection Process ........................................................................................................ 22

4.2 Data Range Flagging ................................................................................................................................... 23

4.2.1 Invalidating Physically Unreasonable Measurements ........................................................................... 23

4.2.2 Wind Speed Standard Deviation ............................................................................................................ 23

4.2.3 Gust Factor ............................................................................................................................................. 23

4.2.4 Wind Vane Standard Deviation .............................................................................................................. 23

4.2.5 Wind Vane Standard Deviation II: Obvious Sensor Degradation ........................................................... 25

4.3 Sensor Relational Flagging ......................................................................................................................... 29

4.4 Data Trend Flagging ................................................................................................................................... 29

4.5 Icing Part I: Obvious Anemometer Stalling and Slow-Down ...................................................................... 29

4.6 Icing Part II: Vane Stalling at Low Temperatures ....................................................................................... 31

4.7 Tower Distortion Effects & Comparative Underperformance ................................................................... 32

4.8 Sensor Integrity .......................................................................................................................................... 34

5 Data Selection ..................................................................................................................................................... 35

6 Post-QC Data Recovery ....................................................................................................................................... 37

6.1 Physical Sensor Recovery Rates ................................................................................................................. 37

6.2 Derived Variable Calculable Rates.............................................................................................................. 41

7 Preliminary Characterization of the Observed Wind Speeds .............................................................................. 43

8 Wind Shear and Vertical Extrapolation ............................................................................................................... 48

8.1 The Approach to Wind Shear ..................................................................................................................... 49

8.2 The Regional Terrain .................................................................................................................................. 50

8.3 Shear Characteristics – All Valid Data between 2012-Oct-05 and 2013-Oct-04 ........................................ 53

8.3.1 Shear Calculated Using Derived Variables UT & UB............................................................................... 54

8.4 Extrapolation to Hub Height ...................................................................................................................... 60

3

9 Turbulence Intensity (TI) ..................................................................................................................................... 64

9.1 Mean and Representative TI Characteristics ............................................................................................. 65

9.2 TI at the Hub Height ................................................................................................................................... 68

10 Climatological Adjustment .................................................................................................................................. 71

10.1 The Long-Term Reference Data Set ........................................................................................................... 71

10.2 Comparison of the Concurrent Target and Raw Reference Data Sets ....................................................... 74

10.3 Quality Control of the Reference Data ....................................................................................................... 77

10.4 Measure-Correlate-Predict (MCP) Process ................................................................................................ 79

10.5 Characterizing the Long-Term Wind Climate ............................................................................................. 82

10.5.1 3-Hour Time-Step MCP Climatological Adjustment Results .............................................................. 86

11 Appendices.......................................................................................................................................................... 89

11.1 Data Recovery Rates By Month and Hour of Day ...................................................................................... 89

11.1.1 Physical Sensors (data to 2013-Dec-12) ............................................................................................ 89

11.1.2 Derived Variables (data to 2013-Dec-12) .......................................................................................... 91

11.2 Wind Roses - Validated Sensor Data .......................................................................................................... 93

11.2.1 Frequency of Wind Occurrence by Sector, and Month (data to 2013-Dec-12) ................................. 93

11.2.2 Frequency of Wind Occurrence by Sector, and Hour (data to 2013-Dec-12).................................... 93

11.2.3 Frequency of Wind Occurrence by Sector, Month, and Hour (data to 2013-Dec-12) ....................... 94

11.3 Wind Shear: The Physical Sensors ............................................................................................................ 102

11.3.1 Shear: South-Southwest-Oriented Sensors U1 & U3 ...................................................................... 102

11.3.2 Shear:North-Northeast-Oriented Sensors U2 & U4 ........................................................................ 106

11.4 Measure-Correlate-Predict Data .............................................................................................................. 110

4

2 Executive Summary

There is growing interest in finding an alternative energy solution to displace diesel generated power in the

Northwest Territories community of Inuvik. Following the Inuvik wind energy pre-feasibility analysis prepared in

March, 2012, by J.P. Pinard and J.F. Maissan1, which suggested that wind energy projects for Inuvik, NWT have the

potential to be developed at a lower cost than diesel2, a wind monitoring campaign was undertaken to measure

the wind resource available in the region. Three specific sites (Inuvik, Caribou Hills, and Storm Hills) were

considered in that pre-feasibility study, and an initial project size of 1.5-1.8 MW was suggested.

Direct measurements were taken near the Storm Hills Distant Early Warning (DEW) Line radar station, 60 km north

and slightly west of Inuvik (Figure 2-1 and Figure 2-3). The meteorological measurements were taken at the New

North Networks communications tower (herein referred to as the Storm Hills met mast) located about 1 km south

of the DEW site and 3 km South of Environment Canada's (EC) Storm Hills climate monitoring station (Figure 2-2

and Figure 2-4).

This wind resource analysis is a comprehensive review of the data collected at the Storm Hills met mast from

October 4, 2012 to March 7, 2014, and is intended to be used as guidance for any proposed wind energy

development in the region. The process and rationale behind data quality control measures are documented in this

report, as are analysis of the results. The assessment does not attach an uncertainty to its findings, though that

would be a logical next step, required in annual energy production analysis. One caveat to the results is that data

recovery rates after quality control were low for most sensors, but reasonable efforts were made to work with the

available data.

1Jean-Paul Pinard, John F. Maissan, Inuvik Wind Energy Pre-Feasibility Analysis (March 28, 2012)

2 Pinard & Maissan

5

Figure 2-1 - Storm Hills met mast relative to Inuvik, and Environment Canada's Trail Valley climate station

Figure 2-2 - Storm Hills met mast relative to Environment Canada's Storm Hills climate station, and the Storm Hills DEW

6

Figure 2-3 - Regional terrain map north of Inuvik, to Storm Hills

Figure 2-4 - Storm Hills terrain map

7

2.1 Analysis Conclusions

The Storm Hills data was recorded at 16.5 m and 39 m above ground level (a.g.l.) and so can reliably be extrapolated to at least 60 m. When correlated with a concurrent data set from the Trail Valley climate station (EC) and back-predicted using historical records, the data implies a reasonably strong 60 m wind regime with an annual average wind speed of around 7.8 m/s. This regime is well-suited to a design-class III wind turbine under the International Electrotechnical Commission (IEC) standard 61400-1. Further extrapolation to 78 m a.g.l. (the hub height of a turbine model known to be suitable for use in cold climates) predicts a mean speed of about 8.1 m/s, and is also suitable to design-class III machines. More importantly, because of the relatively low mean air temperature and low elevation (256 m) at the site, air density is high and so the 50 m a.g.l. mean wind power density is also high, at around 520 W/m

2. This is generally

categorized as class 5 (out of 7), which reflects an excellent wind resource3.

Tower coordinates 68° 53.013' N, 133° 56.893' W Tower base elevation 256 m Anemometer measurement heights 39 m, 16.5 m Measurement period October 4, 2012 - March 7, 2014 Data recovery rate at 39 m 51.4% Observed annual wind speed at 39 m (measured) 7.3 m/s Observed Weibull k (shape) parameter at 39 m 2.05 Observed Weibull c (scale) parameter at 39 m 8.3 m/s Predicted annual wind speed at 60 m (long-term) ~7.8 m/s Weibull k parameter at 60 m 2.19 Weibull c parameter at 60 m 8.7 m/s Predicted annual wind speed at 78 m (long-term) ~8.1 m/s Weibull k parameter at 78 m 2.17 Weibull c parameter at 78 m 9.1 m/s Mean air density 1.314 kg/m

3

Mean turbulence intensity at 15 m/s 0.061 IEC 3rd edition turbine turbulence design class C (site has low turbulence) Mean surface roughness length 0.0001 m Mean power law shear exponent 0.08 Wind power density at 39 m (measured) 505 W/m

2

Wind power density at 60 m (long-term) ~520 W/m2

Wind power density at 78 m (long-term) ~580 W/m2

Wind power class (based on 50m a.g.l.) 5 (of 7 - excellent) IEC 3rd edition turbine design class III

Figure 2-5 - Statistics from the measured data, and the long-term prediction using a correlation with Trail Valley (EC) data

Wind shear, the change in wind speed with height a.g.l., is generally very low. This is as would be expected for a site with few surface roughness elements (such as tall trees or buildings) and a winter snow pack; the mean power law shear exponent was found to be around 0.08 (a typical value would be 0.2). Low shear implies that the selection of a suitable wind turbine may not require the developer to accept the costs associated with purchasing and installing machines with taller hub heights. The typical turbulence intensity (TI) at the site is low, with mean TI of 0.061 under 15 m/s wind conditions (TI-15), while the 90th percentile, representative TI-15 is 0.084. These characteristics describe a turbulence regime suitable to a turbine of IEC 61400-1 turbulence design class C (or A, or B).

3http://rredc.nrel.gov/wind/pubs/atlas/tables/A-8T.html

8

The wind regime is directionally bi-polar (Figure 2-6), with winds most frequently coming from the west-northwest and the east, with a slight southerly component.

Figure 2-6 - Storm Hills measured data wind frequency rose - 1 full year

Figure 2-7 - Long term annual wind speed profile is winter-peaking

Figure 2-8 - Long term diurnal wind speed profile is

afternoon -peaking

9

Figure 2-9 - Modelled regional wind resource map at 78 m a.g.l. (likely turbine hub height). Grid lines represent 10 km.

The wind map in Figure 2-9 was generated in Openwind using a simple wind flow model called NOABL. The model

is driven by a long-term data set at Storm Hills which was created by correlation with data from Environment

Canada's Trail Valley climate station (10 m a.g.l.). The "wind roses" represent the frequency of wind events by

direction sector (16 sectors in all). The EC Trail Valley site is only at 10 m a.g.l., which is close to the ground and so

the wind frequency rose in Figure 2-9 shows heavy influence from frictional drag near the surface, particularly

when wind speeds are low. The roses in Figure 2-10 compare concurrent-period data collected at Storm Hills and

Trail Valley, but only when wind speeds exceed 5 m/s at the respective locations. The wind roses are much more

similar (radial scale is up to 35% frequency of occurrence).

The wind map is generated on a larger scale than would normally be done using a simple flow model driven by a

single met mast, but it shows that there may be sufficient wind resource along the ridges at Caribou Hills to pursue

further wind resource assessment there. Caribou Hills may also be more accessible than Storm Hills, from a project

construction perspective.

10

Figure 2-10 - Wind roses at Storm Hills and Trail Valley when wind speeds exceed 5 m/s (just above turbine cut-in)

11

2.2 Recommendations and Next Steps

Suggestions for any further wind resource campaign at the Storm Hills site include the following (refer to the bulk

of the assessment report for context):

the wind direction vane needs to be replaced as it is damaged - at present all new data is unusable

without corresponding direction data;

consider mounting at least one heated sensor on the tower at the 39 m level to mitigate data losses due

to icing;

boom angles could be changed so that co-levelled sensors are not directly opposite one another (this

helps with quality control procedures);

boom lengths should be extended significantly so as to remove tower distortion affects on the wind

measured (this is critical because there can be unnecessary data loss as a result of booms being too short)

It is the author's opinion that continuing to take measurements at Storm Hills is a good idea so as to provide a

longer-term high-quality data set for future reference; the use of 10 m a.g.l. Environment Canada data sets as

references is not ideal. The recommendations bulleted above will add cost to the existing measurement campaign,

but as a reasonable resource has been indicated, it may prove worthwhile. The addition of a new wind direction

vane is absolutely critical to maintain the present site.

Although erecting a new met tower will be a significant expense, it may be worthwhile from a project-cost

perspective to pursue a concurrent measurement campaign at Caribou Hills (considering little has been spent on

apparatus to-date). A sturdy 50 m winter-rated met tower could be installed in a region free from tall trees and

outfitted with a robust wind resource assessment tower layout:

three sensor levels (20 m, 35 m, 50 m), with boom angles between 45° and 60° about 290°;

six cup anemometers, of which at least one is class 1, and one is heated (requiring a power system);

three wind direction vanes, one at each sensor level, of which one is an RM Young Alpine model;

a pressure sensor;

a relative humidity sensor;

two temperature sensors (top and bottom sensor levels);

and a solar radiation sensor.

The purpose of taking solar and humidity data is to provide information streams to assist turbine manufacturers in

anticipating the likelihood and persistence of ice build-up on their machines; this affects annual energy production

analyses. The purpose of a pressure sensor is to better track air density and therefore better compute wind power

density, and the purpose of having two temperature sensors is to better understand anemometer icing events, as

well as to interpret the any effects owing to summer-time thermal instability.

Wind turbines and cup anemometers are very different things, but there is enough icing and cold temperature

data recorded at the Storm Hills met mast to warrant the consideration of heated blades as a desirable

characteristic for any turbine chosen to be installed in the area around Inuvik. Regardless, the turbine

manufacturers should be engaged to provide information on their most robust cold-weather packages, and a

qualified engineering firm should be consulted to recommend one of the options.

12

Going forward, high-quality long-term data sets for correlation purposes can be purchased from a number of

industry consulting groups; again, the Environment Canada data set can mislead correlation analyses because of

the influence of the ground on their measurements (including temperature).

13

3 Site Description

3.1 General Regional Geography and Climatology

The climate in this region is generally classified as subarctic. According to the 1981-2010 climate data measured at

the Environment Canada climate station Inuvik A (Climate ID 2202570, 66 18’15.000” N, 133 28’58.000” W), the

coldest daily average air temperature was -29.9 C in January and the warmest daily average was 14.1 C in July.

From 1981-2010, the region received 158.6 cm of snowfall on average annually, with heaviest snows in October

(30.1 cm). August is the rainiest month with 36.4 mm of precipitation, on average.

From 1981-2010, measured wind speeds at 10 m above ground level at the Inuvik weather station were highest

during summer (May-August) and the highest monthly average was in June (3.4 m/s). Lowest average monthly

wind speeds were during the winter (December-January, 2.0 m/s average). Throughout the year, wind directions

were most frequently from the east, and the annual average wind speed was 2.6 m/s. These statistics generally

don't match closely with measurements taken at the Storm Hills site.

The Canadian Wind Energy Atlas, based on results from the Environment Canada (EC) Wind Energy Simulation Toolkit (WEST) suggests that wind speeds in the region might resemble those in Figure 3-1

4.

Figure 3-1 – Canadian Wind Energy Atlas: WEST model predicted wind regime near Storm Hills

The following Google Earth imagery has been provided to give a general physical overview of the Storm Hills wind

project site.

4http://www.windatlas.ca/en/index.php

14

Figure 3-2 - Google Earth Imagery of Storm Hills location in North America

Figure 3-3 - Storm Hills location in northern Northwest Territories

15

Figure 3-4 - Google Earth satellite image of Storm Hills site from above

Figure 3-5 - Google Earth image of the site, looking north

16

Figure 3-6 - Google Earth image of the site, looking east

Figure 3-7 - Google Earth image of the site, looking south

17

Figure 3-8 - Google Earth image of the site, looking west

18

Figure 3-9 - Green indicates where at least 50% of the vegetation is taller than 2 m in height - Canadian Vector Database

19

Figure 3-10 - The Storm Hills meteorological mast at the New North Networks communications station, looking southeast

20

3.2 The Data Set

The raw logger data was scaled and interpreted as per the instrument and tower layout specifications described in the site installation log, and by using the NRG Symphonie Data Retriever 7.03.15 software. As with any field data, the sensor measurements inevitably contained faulty or questionable readings that required special scrutiny to determine what was to be filtered out prior to the final site resource analysis. A number of industry-standard data flagging routines

5 were modified to suit the site and applied to the data using

the Windographer 3.2.5analysis software. Final data removal was completed by visual inspection and interpretation by the analyst, taking into consideration the flagging. Reasonable leniency was given in some cases to retain useful data recovery rates without drastically affecting the final analysis.

3.3 Sensor Models and the Data Format The meteorological tower at Storm Hills is a lattice mast with a 2’ distance between the legs. It is equipped with anemometer pairs at two altitudes, 39m and 16.5 m, as well as a vane at 39 m and a temperature sensor 2 m off the ground. 39m up the tower is one NRG Class-1 cup anemometer mounted on a boom, 40“ away from the nearest lattice leg. The mounting boom is pointing north-northeast, away from the tower. Pairing that cup is an R.M. Young Alpine Wind Monitor horizontal-propeller anemometer. The propeller sensor is also the wind direction vane and is mounted 40" from a mast leg, on the same boom, which is oriented towards the south-southwest with the vane dead-band aligned with true north. Figure 3-10 is a view looking southeast towards the Storm Hills station, with the building to the west of the tower and the booms set so that the instruments are relatively clear of the existing communications equipment. There are two NRG Class-1 cup anemometers at 16.5 m with similar bearings and boom mountings to the two sensors at 39 m.

Variable Altitude

(m) Sensor Type Model

Boom Orientatio

n

Boom Length (")

Response Offset Response Slope

(scale factor)

U1 39 propeller

anemometer R.M.Young

Alpine ~SSW 40 0 m/s cut-in 0.098 m/s/Hz

U2 39 cup anemometer NRG Class 1 ~NNE 40 0.2 m/s cut-in 0.766 m/s/Hz

U3 16.5 cup anemometer NRG Class 1 ~SSW 40 0.18 m/s cut-in 0.768 m/s/Hz

U4 16.5 cup anemometer NRG Class 1 ~NNE 40 0.21 m/s cut-in 0.766 m/s/Hz

D1 39 wind vane R.M. Young

Alpine ~SSW 40

0° (true) dead-band alignment

0.351

T1 2 temperature sensor NRG 110S - - -86.383 °C 0.136

Table 1– Summary of the sensors and variable naming convention; The R.M. Young Alpine Wind Monitor vane has a response threshold of 1.1 m/s

The data was collected in a raw format by an NRG Symphonie logger, which samples sensor signals once every two seconds and records statistical values (average, standard deviation, maximum and minimum) over ten-minute time steps.

5 AWS Truepower, Wind Resource Assessment Handbook: Final report (Albany, New York, USA: State of New York

Energy Research and Development Authority, 2010) Section 9 – Data Validation, 9-1 to 9-11

21

This averaging period is the industry standard as it minimizes remote data storage requirements while maximizing relevant data collection within the boundary between mesoscale (1 to 100 km) and microscale (< 1 km) weather events. Mesoscale meteorological phenomena, like cyclones, pass over minutes to days, while microscale phenomena, like turbulence or gusts, last seconds to minutes

6. This averaging period represents part of an energy

gap in the standard near-to-ground wind intensity spectrum over which speeds can be studied statistically because of the low prevalence of noise due to atmospheric eddies

7.Those eddies which do arise on this time scale are the

larger of the micro-scale turbulent ones which are important because they impact wind turbine performance and can cause gradual wear and tear on the machines.10 minutes is also a time scale relevant to energy off-takers like power utilities. The recorded variables in this analysis are referred to as: Ui, Ui-SD, Ui-max, and Ui-min for the individual anemometers; D1 and D1-SDfor the direction vane; and T1 for the temperature sensor. The full data collection period was 2012-Oct-04 to 2014-Mar-07.

6Janardan S. Rohatgi & Vaughn Nelson, Wind Characteristics: An Analysis For The Generation Of Wind Power

(Canyon, Texas, USA: Alternative Energy Institute, West Texas A&M University, 1994) 11 7Roland B. Stull, An Introduction to Boundary Layer Meteorology (Dordrecht, The Netherlands: Lower Academic

Publishers, 1988) 32-33

22

4 Quality Control (QC)

4.1 Data Flagging and Inspection Process

The following is a generalized description of the process used to exclude data or flag it for special scrutiny.

1. Apply data range flags8:

Data records outside the reasonable site-specific ranges for each sensor variable and their important derivatives were flagged for later scrutiny.

2. Apply sensor relational flags

9:

A sequence of standard criteria examining the relationships between the different sensors was used to flag data for later scrutiny.

3. Apply data trend flags

10:

A series of standard criteria examining how the data changed over given periods were used to flag for later scrutiny.

4. Remove obvious icing events:

a. Anemometer data revealing obvious stalling or slow-down due to icing was excluded using

screening tests and manual examination of the data as a time-series.

b. Vane data revealing obvious stalling due to icing was excluded using screening tests and manual examination of the data as a time-series. All corresponding anemometer data was also removed.

5. Examine tower distortion:

a. Ratios and differences between co-levelled anemometer data were plotted against wind

direction to infer the normal wind speed distortion characteristics of the met mast. b. Remove tower shading:

Data was excluded when the wind direction likely placed anemometers within the met mast's wind shadow. Slower co-levelled anemometer data was also discarded.

c. Flag acceleration zones:

Data was flagged when the wind direction implied anemometers might have experienced higher-than free-stream wind speeds as a result of the presence of the met mast.

6. Detect sensor underperformance:

All flags from steps 1-3 and 5c were used to identify and manually exclude data from underperforming sensors. This was done by visual inspection of the time-series data.

8 AWS Truepower, Wind Resource Assessment Handbook: Final report, 9-4

9 AWS Truepower, Wind Resource Assessment Handbook: Final report, 9-5

10 AWS Truepower, Wind Resource Assessment Handbook: Final report, 9-5

23

4.2 Data Range Flagging

4.2.1 Invalidating Physically Unreasonable Measurements

Wind speed Ui and maximum 2-second gust Ui-max data for all anemometers fell within the reasonable ranges.

The Ui data was all between the sensor offset (cut-in wind speed) and 30 m/s. All 2-second gusts fell within the

range spanning from the offsets to 35 m/s.

Wind direction D1data were within the 0° to 359° range, as referenced to true north (the dead-band alignment).

TemperatureT1data were within the range -37.4°C to 30.1°C, which is reasonable for an arctic environment over the course of a full year or longer.

4.2.2 Wind Speed Standard Deviation

The standard deviation of the wind speed Ui-SD of a 10-minute anemometer data step was flagged if it exceeded

3.0 m/s. These events typically happen when an anemometer is freezing or thawing at the periphery of an icing

event, or during events of severe underperformance.

4.2.3 Gust Factor

The ratio of the maximum recorded 2-second wind gust to the 10-minute average wind speed is the gust factor11

:

Anemometer data segments containing time steps with gust factors exceeding 2.5 were flagged12

. Time steps

reporting extreme gust factors often occur at the beginning or end of anemometer icing periods.

4.2.4 Wind Vane Standard Deviation

It was important to determine the normal performance characteristics for the wind vane at the site. The standard

deviation of wind direction D1-SDshould fall within reasonable ranges, but those ranges are site-specific. When D1-

SD was very low there was a good chance the vane was partly frozen in place or was not responding correctly

because the wind speeds were low, and perhaps below the sensor threshold; the R.M. Young Alpine Wind Monitor

vane has a threshold sensitivity of 1.1 m/s. Stalling due to icing is dealt with later, so this section deals with

excessively high D1-S Dvalues.

Elevated D1-SD can indicate the passage of a front, where changes in wind direction and temperature can be

sudden and genuinely erratic. It can also be indicative of a lull, where low wind speeds fail to meet the vane

threshold, producing an intermittent response to the wind direction. High D1-SD is indicative of unusually erratic

behaviour in a wind vane if it occurs for an extended period of time. Erratic vane behaviour can have a number of

different causes, which may include damage or icing-related obstruction to the sensor. Typically vane standard

deviation values require scrutiny when they exceed 75° in a single time step13

, though that measure is site specific.

11

Windographer 3.2.5 documentation: Gust Factor 12

AWS Truepower, Wind Resource Assessment Handbook: Final report, section 9-5 13

AWS Truepower, Wind Resource Assessment Handbook: Final report, section 9-4

24

Figure 4-1- log (D1-SD) vs. D1 before and after full quality control Coloured by T1

Figure 4-2 - log (D1-SD) vs. U1 before and after full quality control

Coloured by T1

For this site D1-SDdata was initially flagged based on the left-hand frames of Figure 4-1 and Figure 4-2, then

eliminated by visual inspection of the time-series. Vane data was eliminated when reporting extremeD1-SDif it was

felt that the actual wind direction could not reasonably be known. All anemometer data was also eliminated in

these events under the assumption that the R.M. Young vane was the most robust sensor on the tower. Inspection

of the data set confirmed this was not too conservative an approach.

Figure 4-3shows a period during which the wind vane was erratic and U1 was experiencing an icing event.

25

Figure 4-3 - High D1-SD and extended erratic vane behaviour with anemometers also flagged

4.2.5 Wind Vane Standard Deviation II: Obvious Sensor Degradation

Figure 4-4 - D1-SD above, log(D1-SD) below, withT1, all prior to quality control

Erratic D1-SD time steps roughly marked in green

On 2013-Dec-12anemometer U1 was stalled in an icing event. From that point until the end of the data set on

2014-Dec-12, the wind vane D1behaved in an abnormal and erratic fashion. When compared to the same time

period a year earlier, and to the majority of the data set, the vane was biased northwards and behaved erratically,

frequently reporting excessively high D1-SD values when not pointing northwards.

26

D1 - Affected

2013-Dec-12 to 2014-Mar-07

D1 - Same period the previous year

2012-Dec-12 to 2013-Mar-07

D1 - The full unaffected data set

2012-Oct-04 to 2013-Dec-12

EC Trail Valley

2013-Dec-12 to 2014-Mar-07

EC Trail Valley

Mean January 2000-2014

EC Trail Valley

Mean February 2000-2014

Figure 4-5 - Wind direction frequency of occurrence prior to quality control

A northerly wind regime between December and March is inconsistent with both historical data at the tower and

with reference data from a nearby Environment Canada station at Trail Valley, where there is a sonic anemometer

mounted 10 m above the ground. Figure 4-5illustrates the northerly bias issue comparatively, using wind

frequency roses over time periods of interest.

27

Affected - log(D1-SD) vs. D1

2013-Dec-12 to 2014-Mar-07

log(D1-SD) vs. D1

2012-Dec-12 to 2013-Mar-07

log(D1-SD) vs. D1

2012-Oct-04 to 2013-Dec-12

Affected - Mean of D1-SD vs. D1

2013-Dec-12 to 2014-Mar-07

Mean of D1-SDvs. D1

2012-Dec-12 to 2013-Mar-07

Mean of D1-SDvs. D1

2012-Oct-04 to 2013-Dec-12

Figure 4-6 - Prior to quality control

Left Panels: D1-SD frequently high, especially when non-northerly Center Panels: D1-SDsame period the previous year

Right Panels: D1-SD the full unaffected data set

After 2013-Dec-12, D1-SD was reported too frequently outside the reasonable range of values. Earlier data

showed the value was consistently between 0.5° and 10° for a typical time step. Figure 4-6shows how D1-SD varied

across relevant time periods and the top-left panel demonstrates that after 2013-Dec-12 the vane response

characteristics were unusually erratic and biased northwards. The comparison is made to both: the same period

one year earlier (center panels); and to all of the prior data (right panels). The vane data suggests that abnormal

vane behaviour may have begun to occur as early as 2013-Dec-05.

All the sensor data after 22:40 on 2013-Dec-12 was marked as invalid and excluded from the analysis, because

without vane data a full quality control regime could not be implemented.

Figure 4-7 - After 2013-Dec-12, while D1 was erratic, U1consistentlyunderperformedand was flagged invalid

28

The issue with the R.M. Young anemometer may have extended further back in the data

set.U1beganperformingpoorly when compared toU2after coming out of a 24-hour period of icing-related slow-

down on 2013-Sep-24. It was also underperforming when compared to U3, which was directly below it, even while

U3 performed acceptably against U4. For the majority of the affected time period the winds were from a direction

in which U2 was expected to experience tower-induced flow acceleration (from the free-stream wind speed). This

complicated the analysis; however examination of the time-series data suggested that U1 was indeed

underperforming against U2 in an uncharacteristic fashion which was more than just a result of tower distortion.

See Figure 4-8 and Figure 4-9.

Figure 4-8 - U1 vs. U2

2013-Sep-23 to 2013-Dec-12 U1consistently underperforming against U2

Figure 4-9 - U1/U2vs. D1

2013-Sep-23 to 2013-Dec-12 Tower distortion signal exacerbated by U1

underperformance

All of the U1 data after 2013-Sep-24 was labelled as subject to the effects of icing and therefore excluded from the

analysis, though it was used to help flag obvious icing-related underperformance in U2 before being discarded.

Assumed damage to the R.M. Young wind direction sensor resulted in a recovery loss of 16.2% of the original U1

and D1 data sets. U1 probably suffered greater losses in relation to this damage, though that was removed later as

possibly icing-related.

As of the final data segment available (2014-Mar-07), the R.M. Young sensor was unreliable in its functions as both

a propeller anemometer and as a wind direction vane. It is recommended that it be replaced.

29

4.3 Sensor Relational Flagging

A set of standard criteria was employed to flag data showing odd comparative values between co-levelled

anemometers, and between measurements at different heights. The purpose of these flags was to indicate periods

likely to be reporting: underperformance due to icing; excessive and incorrect shear values; and acceleration or

slow-down of the free-stream wind speed due to the presence of the met mast.

As will be discussed shortly, the tower configuration suffered heavily from tower distortion effects and icing. This

rendered blanket tests between co-levelled anemometers relatively ineffective as the distortion was sector-wise

ubiquitous and extremely non-uniform.

Climate-related anemometer slow-down and stalling rendered attempts to verify characteristic shear scenarios

was difficult as there were relatively few time steps where both anemometers at 16.5 m could be used to confirm

the actual free-stream wind speed at that level. The site itself should have low shear, being an arctic tundra

location with a winter snow blanket at the ground, however there were several instances where relatively high

shear scenarios could neither be confirmed reasonably valid, nor rejected as impossible. There was no sensor at

the 16.5 m level which could be relied upon as being the primary anemometer under all circumstances.

4.4 Data Trend Flagging

Standard anemometer 1-hour performance trend tests ultimately revealed nothing unusual which was not later

flagged for some other, more-obvious reason. Anemometer data segments flagged for sudden changes were found

to be consistent across the set of sensors and so were considered legitimate phenomena, often occurring at high

wind speeds and concurrent with significant valid changes in wind direction.

Temperature sensor data flagged by 1-hour change tests were considered reasonable for an arctic climate,

happening primarily in the warmer months, early in the morning and late in the evening, and often coinciding with

sudden and significant changes in wind direction and speed. These events were most likely related to the diurnal

solar cycle.

4.5 Icing Part I: Obvious Anemometer Stalling and Slow-Down

Some standard icing-related tests were used to flag the data, then the full anemometer time-series was examined

manually and segments were excluded from the analysis when determined to be experiencing extended periods of

stalling and/or obvious slow-down as a result of variable-temperature and cold-temperature events. Co-levelled

anemometers were examined independently, with all statistical variables considered (Ui ,Ui-SD , Ui-max , Ui-min)

for each time step. D1, D1-SDandT1 data was also used to determine meteorological conditions when periods of

stalling were found. Anemometer data determined independently of other sensors to be affected by obvious icing

was excluded from the analysis.

30

Figure 4-10 - Obvious icing-related stalling and slow-down flagged in U1 (top) and U2 (middle)

U, Ui-SD, Ui-max, Ui-min used to identify anemometer icing D1, D1-SD, T1 used to identify changing weather conditions

31

4.6 Icing Part II: Vane Stalling at Low Temperatures

The wind vane data time-series was examined and flagged as being likely affected by icing when the sensor

experienced extended periods of flat-lining, withD1-SDat or near zero, and with T1near or below 1°C or changing

significantly. Short data periods (one time step, or so) with similar characteristics were ignored if it appeared that

the vane data was consistent with what would have been expected at that time, especially if the anemometers

were reading reasonable wind speeds.

The vane was a robust alpine model designed for harsh winter climates. Whenever the vane was affected by icing

all the anemometers were also assumed to be affected and the data was excluded from the analysis. This

conservative approach was found to be consistent with the time-series data.

Figure 4-11 - Obvious case of an extended period of D1 icing and corresponding anemometer stalling

Top: all anemometers stalled or slow (was also independently flagged) Middle: D1 flat-line indicating vane stalling, T1 low

Bottom: D1-SDvery low

32

4.7 Tower Distortion Effects & Comparative Underperformance

The presence of the meteorological mast influences wind speeds around the tower which can cause an

anemometer to encounter a wind speed other than the actual free stream one. This effect is described in the IEC

documentation14

and is dependent on the type of mast, length of the anemometer mounting boom, and the wind

direction relative to the mast and sensor.

An anemometer downwind of the mast is effectively shaded and reads below the actual free stream wind speed,

often also reporting increased turbulence intensity. To a lesser degree a similar slow-down is measured when an

anemometer is mounted upwind of the mast. Tower shading can be graphically examined by polar-plotting the

ratio and/or difference of co-levelled sensors against the corresponding wind vane direction readings.

Figure 4-12 - Tower Distortion Polar Plots

Top: U1/U2vs.D1 scatter polar plots prior to and after data validation; and statistically, after data validation Bottom: U3/U4vs. D1scatter and statistical polar plots

Sectors of higher-than free-stream wind speed marked: U1 and U3 in blue, U2 and U4in green

Figure 4-12plots the wind speed ratios U1/U2 and U3/U4 vs. D1 after: only obvious icing data has been removed

(left); and after full quality control (centre and right). The shading of anemometers U1 and U3 is clear when the

winds are from the north-northeast, and the shading of U2 and U4 occurs from the south-southwest.

There remained considerable scatter in the representations of the data sets where only obvious icing had been

removed (left panels). Tower shading is clear, but most of the noise was a result of co-levelled anemometers

underperforming relative to one another in the severe arctic conditions. That faulty data was excluded by:

examination of the data streams together in a time-series, making reference to flags and looking for clear co-

levelled anemometer relative underperformance or excessive shear between sensor levels; and by examination of

various scatter plot presentations of the data to determine general validation patterns.

14

IEC 64100-121-1, Annex G.6, 70

33

Absolute sensor underperformance, in a general sense, was suspected but difficult to quantify as none of the

sensors was deemed of high enough quality to be taken as a benchmark for comparison.

Once the data showing obvious icing, tower shading, relative underperformance, or excessive shear were excluded

from the analysis, the data scatter factor of the wind speed ratio vs. wind direction polar plot at the 39 m level

changed from 0.321 to 0.014.This factor is the weighted mean of the standard deviation of the wind speed ratios

when binned into 72 direction sectors15

. At the 16.5 m level, the scatter factor dropped from 0.140 to 0.010. At

both levels these changes are indicative of a significant improvement in the general agreement in wind speed

readings between co-levelled sensors for data time steps included in the final analysis. However, in both cases the

percent of possible data time steps with valid paired readings and concurrent valid vane data dropped significantly:

58.1% to 24.3% at the 39 m level; and 49.7% to 32.4% at the 16.5 m level.

The plots in Figure 4-12 demonstrate significant tower-induced distortion of the free-stream wind speed in one of

the anemometers at both sensor heights, almost all the way around the met mast. In the case of the 39 m

anemometers, the two sensors are different makes and models so such a ratio could be expected to show some

deviation from unity, however in this case the distortion is extreme.

The free-stream wind speed distortion appears primarily because the sensors were mounted on short

anemometer booms on a lattice-type tower. Ideally the speed ratio plots in Figure 4-12would cluster around the

unity ratio circle, but the tower distortion factors were high: 0.041 at 39 m; and 0.032 at 16.5 m. These distortion

factors quantify the weighted non-unity of the median values of the wind speed ratios when binned by sector16

.

Ideally, met mast booms should be long enough and oriented in such a way as to ensure the horizontal wind speeds encountered by a sensor are within 99.5% to 100.5% of the actual free-stream speed during the most frequently occurring events, as well as during the most energetic wind events (if possible). For a three-legged lattice mast the wind speed near to the tower can be expected to deviate from the free-stream value as described in the IEC documentation

17. Effects seen in the measured wind speed due to upwind speed deficits or around-

tower acceleration are exacerbated the closer the sensor is mounted to the tower. Interpretation of the documentation suggests the value of R/L, where R is the distance from the geometric center of the tower to the anemometer and L is the distance between the lattice legs, should be higher than 5.0 in order for the sensors to experience wind speeds that are at least 99% of the free stream wind speed when sited upwind of the mast. The value of R/L for this met mast configuration was around 1.5 (low) for all anemometers. Distortion on this mast was fairly high for that reason; severe deficits occurred upwind of the mast, and beyond-free-stream speed-ups were seen at anemometer locations around the mast from the upwind direction.

In Figure 4-12the regions of tower distortion have been marked; the blue ovals highlight wind directions where the odd numbered sensors, mounted to the south-southwest of the tower, measured higher than free-stream wind speeds, while the green circles highlight the same for the even numbered sensors mounted north-northeast. Very little of the data in this data set was free from tower distortion effects.

15

Windographer 3.2.5 documentation: Scatter Factor, ∑ ∑

; is the number of direction

sectors in the polar plot; is the standard deviation of the sensor speed ratios in sector ; is the number of records in sector 16

Windographer 3.2.5 documentation: Tower Distortion Factor, ∑ | | ∑

; is the number

of direction sectors in the polar plot; is the median value of the ratios of wind speeds in sector ; is the number of records in sector 17

IEC 64100-121-1, G.6.2

34

4.8 Sensor Integrity

Figure 4-13 - U1 vs. U2 linear best fit

8 direction sector bin centres: # valid data pairs; best-fit slope m = U1/U2; R^2 Overlapping 6 month sample periods

Figure 4-14 - U3 vs. U4 linear best fit

8 direction sector bin centres: # valid data pairs; best-fit slope m = U3/U4; R^2 Overlapping 6 month sample periods

The anemometers on this met tower, being unheated and subjected to arctic conditions had the potential to

become damaged or degraded by icing wear-and-tear, even over the brief period of data collection. As pointed out

in the quality control, U1 and D1 were affected, possibly in this manner, as of 2013-Dec-12.

To determine if there was any evidence of subtle sensor degradation in the quality controlled data, the time steps

containing validated co-levelled anemometer pairs were divided into overlapping 6-month segments and then

further separated by eight wind direction sectors to account for tower distortion variations in the relative sensor

responses.

Figure 4-13 and Figure 4-14 show the relative performance between U1 and U2 was consistent over the validated

dataset, and the same can be said betweenU3 and U4. For all anemometers the sensor performance did not

degrade noticeably over the period of validated data prior to 2013-Dec-12, after which the data could not be

binned by direction sector as D1 was invalidated.

35

5 Data Selection One common practice in wind resource assessment is to mount co-levelled anemometers on the meteorological mast so that neither sensor need be assumed to be the primary one. This is done by using general knowledge of the regional wind regime to guide in positioning sensor booms in adherence with the IEC orientation recommendations

18.

However, if the sensors are deemed to be of significantly different accuracy or reliability, then a primary might still be identified. In that case data from the secondary sensor would be used to substitute into the primary data set when required. In this case it was not possible to strictly follow the IEC standards, owing to resource and time constraints, and the ready availability of other infrastructure. It should be noted that the mounting mast was not a standard meteorological mast, but rather a communications station equipped with multiple transmission devices, the effects of which have not been examined in this report. At this station, and in keeping with good practice, the top measurement level contained at least one class-1 anemometer, as well as a robust R.M. Young winter-weather sensor which doubled as a high-quality vane. In this scenario, given the 180° boom separation and the expected wind regime, the NRG cup anemometer might be selected as the primary sensor because of its independent MEASNET calibration and general certification as a class-1A sensor designed for industrial wind resource assessment. In that case the R.M. Young would be deemed as the secondary as its calibration is simply factory-specified and, being a propeller sensor, it requires that the vane be functioning properly in order to measure the correct upwind speed. Propeller sensors also differ from cup sensors in the degree to which they are sensitive to turbulence; turbulence intensity is an important quantity in wind resource assessment. In an arctic climate and given the data recovery statistics (section 6) from the months it was functioning properly, the R.M. Young sensor was determined to be the more reliable sensor after initial quality control. Still, in this assessment neither anemometer was deemed to be the primary because of the prevalence of tower distortion effects from all but the most the most frequent wind directions.

Figure 5-1 – Winds were of the highest frequency from directions least affected by tower distortion Distortion was more severe at the 39 m level for southerly and westerly directions and more severe at

18

IEC 64100-121-1, Annex G.6.2, 72

36

16.5 m for northerly and easterly directions Overall Distortion factors were0.041 at 39 m, and 0.032 at 16.5 m

All data recorded after 2013-Dec-12 was eliminated from the data set because a proper quality control regime could not be performed without vane data; tower distortion analysis was impossible. As well, taking into consideration the severe distortion described in charts like Figure 5-1, all anemometer readings were removed from the analysis if the charts suggested the sensors were likely to be experiencing higher-than free-stream wind speeds due to tower distortion effects. In sectors where data from both sensors remained for any time step, the average value reported by the two sensors was taken as the free-stream wind speed. The final data set at each sensor level used in the preliminary assessment (Section 7) consisted of: in most sectors, the validated wind speeds measured by the sensors most likely to be exposed to the free-stream wind; and in some sectors, the average wind speed where both sensors reported valid data. The latter were also the sectors with the highest frequency of wind events. The variables used to label these derived data sets were called UT and UB for the 39 m and 16.5 m sensor levels respectively. The final selected data sets used in the analysis are displayed graphically in Figure 5-2.

Figure 5-2- All validated wind data, 2012-10-04 to 2013-12-11 Top: U1 vs. D1; U2 vs. D1

Bottom: U3 vs. D1; U4 vs. D1

Of the 62,931 time-steps prior to 2013-Dec-12, there were 1,746 where both U1 and U2 reported valid data and 2,157 where both U3 and U4 were validated. The total number of data points recovered for each sensor were: U1 31,010; U2 8,988; U3 18,129; U4 16,861. The derived variables UT and UB recovered 38,252, and 32,833 data points respectively. Full data recovery statistics are in section 6.

37

6 Post-QC Data Recovery

6.1 Physical Sensor Recovery Rates

Figure 6-1 - Post-QC data recovery rates (%) by month and year over the full collection period

Top: Vane D1 (blue); anemometers U1 (red), U2 (orange), U3 (yellow), U4 (green) Bottom: Heat Map of all physical sensor recovery statistics

The recovery rate for data from T1, the NRG 110S temperature sensor, was 100% over the course of collection.

Until it began behaving in an erratic fashion with a northerly-bias, recovery rates for the R.M. Young alpine wind

vane were excellent through most months; it was the best performing wind sensor while active.

Data collected after 2013-Dec-12 was eliminated because of the absence of vane data. Otherwise, recovery rates

varied significantly from season-to-season for all anemometers, with the deepest winter months from November

38

to March being the worst collection periods. This was particularly true for sensors U2, U3, and U4. All the

anemometers had data excluded for two primary reasons: stalling or underperformance likely related to icing; and

because preferential treatment was given to sensors more likely to be in the free-stream wind flow (i.e. not

affected by tower distortion effects).

U1, the R.M. Young sensor, performed well against all three of the NRG Class-1 cup sensors until it became

damaged in 2013-Sept and was removed from the analysis.

Generally, when reporting the mean wind speed read by any one sensor over a one month period, that sensor

should have a minimum data recovery rate of 80% for that month. In this data set none of the anemometers

experienced a full calendar month in which the recovery rate could reliably be said to meet that threshold, and

rarely was a threshold of 70% was met. For that reason, where deemed necessary, this wind resource assessment

will quote the UT and UB variables derived from combinations of the valid sensor data, and results from a

correlation of those with a long-term reference data set.

Figure 6-2shows that anemometer validation rates were high in wind direction sectors that had the greatest

number of occurrences, which was expected given the boom orientation. The low recovery rates for anemometers

U1 and U3 when the wind was from the north-east is primarily due to tower shading and distortion, and the same

is true for U2 and U4 when the wind was from the south-west.

The heat chart in Figure 6-3 demonstrates the data recovery rates for anemometers U1 and U3, the south-

southwest mounted sensors, tended to be higher in the mid-morning to early-afternoon hours than they were

overnight. The opposite was true for U2 and U4, the north-northeast mounted sensors. This phenomenon was

most prevalent during the summer months (Appendix 11.1.1, Figure 11-1 to Figure 11-4) and appears to be a result

of diurnal trends in wind direction and the associated tower shading and distortion validation rules. During the

summer months the vane reported more frequent north and north-easterly winds overnight (Figure 6-5), dropping

the overall recovery rates for the south-southwest-mounted sensors and boosting them for the north-northeast

sensors. This may be a summer nocturnal jet.

Figure 6-2 - Anemometer validation rates (left) and number of valid occurrences (right), by wind sector

U1 (red), U2 (orange), U3 (yellow), U4 (green)

39

Figure 6-3 – Mean anemometer hourly recovery rates

Figure 6-4

Left: Sensor recovery rates D1 (blue), U1 (red), U2 (orange), U3 (yellow), U4 (green), by T1 temperature Right: Wind frequency rose for T1 > 20°C

The left graphic in Figure 6-4 shows that sensor recovery rates for D1 and the anemometers U1 and U3 all

generally increased with temperature. U2 and U4, the north-northeast mounted sensors had weak recovery rates

at temperatures above 20°C. When temperatures were warm the winds were generally southerly or south-

westerly (Figure 6-4, right), neither of which occurred very often, but when they did U2 and U4 were shaded or

distorted by the tower.

It's also clear that U2, U3 and U4 did not generally perform as well as U1 at low temperatures. This is likely owing

to the sensor make and model; the R.M. Young Alpine sensor is a more robust and winter-hardy sensor than is the

NRG Class-1 cup sensor, though the precision of the NRG is theoretically superior when conditions are favourable.

40

Figure 6-5 – Top: Hourly recovery Rates for U3 (top left) and U4 (top right), for 2013-Aug

Bottom: Bi-hourly wind frequency roses for 2103-Aug

41

6.2 Derived Variable Calculable Rates

Figure 6-6 - Post-QC derived variable calculable rates by month of data collection

The derived variables UT and UB are the validated wind speeds reported on in section 7of this wind resource assessment, defining the resource at the 39 m and 16.5 m sensor levels respectively. The derived variables UT and UB are closer to the free-stream wind speed than are any of the actual physical

sensor measurements because of the prevalence of tower distortion effects. Both derived variables have useful

data return rates by wind sector (Figure 6-8)and typically had higher rates after noon (Figure 6-8), but the trend

was subtle and wasn't consistent across months (Appendix 11.1.2).

As expected, the derived variables had high calculable rates when the temperature was higher. UB had weaker

recovery rates than UT at low temperatures as the R.M. Young, mounted at 39 m, was the hardiest sensor model

(Figure 6-9).

42

Figure 6-7 - Derived variables: calculable rates (left) and number of valid occurrences (right), by wind sector UT(blue), UB (purple)

Figure 6-8 - Mean derived variable hourly

recovery rates

Figure 6-9 - Derived variables calculable rates by temperature

43

7 Preliminary Characterization of the Observed Wind Speeds

The following wind speed characterizations use the UT and UB data sets. Note that they are fragmented, as shown

in Section 6.2, and are biased against low wind speed events coinciding with sensor icing flagged in the quality

control process. Of greater interest will be the statistics quoted in section 10on long-term climatological

adjustment of the extrapolated hub-height data set.

Figure 7-1 - Preliminary measured wind summary for derived variables UT and UB Left: Entire collection period, 2012-10-04 to 2014-03-07 (winter-biased)

Right: One full year with good overall recovery statistics, 2012-10-05 to 2013-10-04

44

Figure 7-2 - UT and UB monthly wind speed means over the entire data collection period

Recovery threshold for averaging was 60%, 50%, and 40% for each month (top to bottom); Typical one-month recovery threshold is at least 80% for averaging

45

Figure 7-3 - Monthly mean diurnal wind speed profiles, 2012-10-5 to 2013-10-04

UT blue, UB purple Note that recovery rates for UB were very low for 2012-Dec and 2013-Mar

Figure 7-4 - UT wind speed probability distribution, 2012-10-5 to 2013-10-04

Weibull parameters: shape, k = 2.05; scale c = 8.27

46

Figure 7-5 - UB wind speed probability distribution, 2012-10-5 to 2013-10-04

Weibull parameters: shape, k = 1.97; scale c = 7.46

Figure 7-6 - Wind frequency roses, 2012-10-5 to 2013-10-04

UT blue, UB purple

47

Figure 7-7 - Mean wind speed roses, 2012-10-5 to 2013-10-04

UT blue, UB purple

Figure 7-8 - Proportion of total wind energy, 2012-10-5 to 2013-10-04

UT blue, UB purple

48

8 Wind Shear and Vertical Extrapolation Wind shear refers to the change in horizontal wind speed with altitude. Typically this results as momentum is transferred downwards from the atmosphere into the ground due to friction between the moving air mass and roughness elements at the surface. In this usual situation the wind speed increases with height above the ground. In wind regimes with greater shear characteristics wind turbines experience more wear and tear owing to mechanical loading. In cases of high shear wind turbines may need to be installed with higher hub heights, even in strong wind regimes, in order to reduce shear-induced loading. The shearing effect of surface roughness elements is described quantitatively by a measure known as the roughness length, z0. For an arctic tundra site with short grassy vegetation in the summer months and blanket snow cover in the winter months the site characteristic roughness was expected to be around 0.001, which is low

19 and could be characterized as roughness class 0 (zero), which can qualitatively be described as smooth

20.

Roughness length is computed from wind speeds measured at different altitudes, and if known can be used to extrapolate the wind to another altitude of interest z via the empirical logarithmic wind profile

21:

( )

( ) (

)

(

)

Where zr is an altitude at which a reference wind speed is known. When characterizing the wind shear at a site, it is most common to quote the 'shear exponent' in the so-called power-law wind shear profile, commonly referred to as the shear exponent α

22:

( ) ( ) ( )

19

http://en.wikipedia.org/wiki/Roughness_length 20

http://130.226.56.171/Support/FAQ/WebHelp/TableofRoughnessLengths.htm 21

Rohatgi& Nelson 42. 22

Rohatgi& Nelson, 43.

49

8.1 The Approach to Wind Shear

Though this site has very low wind shear in general, there is evidence of a (likely) stability-related nocturnal increase in wind shear during the summer months when the atmosphere is more affected by diurnal solar heating patterns between May and October. It’s possible this is part of a nocturnal jet and there is further suggestion of this in the summer diurnal wind frequency roses in Appendix 11.2.3.

Figure 8-1 - UT and UB in June 2013, with higher nocturnal shear evident

Characteristic shear values used in hub-height calculations are computed using statistics derived from individual time step calculations in such a way that maximizes the use of validated data, while at the same time minimizing the effect of tower distortion on the results. The two derived wind speed data sets UT and UB (as displayed in Figure 8-1) would not typically be used to derive the primary shear values used in computing the hub-height wind speed, which is discussed in Section 8.4. Those data sets, being comprised of time-steps with mixed sensor orientations, may introduce increased uncertainty into what are already sensitive and assumptive shear calculations. The data selection process outlined in section5 attempted to account for the effects of tower distortion on the wind speeds reported for each sensor level. Typically, time step shear values are computed employing coincident data from the actual un-blended anemometer data sets (i.e. U1, U2, U3, and U4). By this standard the shear calculations to hub height are carried out between sensors of similar boom orientations so as to reduce tower distortion influences on the shear uncertainty

23. However, at this site the prevalence of tower distortion in the data set meant that the quality

control process eliminated large portions of the data from each sector, reducing overlapping from oppositely-oriented booms on the tower. Essentially the derived data set is comprised of data from co-oriented sensors at the two different tower heights, so the UT and UB variables were acceptable for use in calculating shear and extrapolating wind speeds to the 60 m hub height, 1.5 times the highest sensor, which is the reasonable limit for shear extrapolations.

23

AWS Truepower, Wind Resource Assessment Handbook: Final report, 10-6

50

8.2 The Regional Terrain

Figure 8-2 - Terrain Maps of Storm Hills Site

51

Figure 8-3 - D1 Wind frequency rose

The vane frequency rose in Figure 8-3shows the majority of the wind events were from west-northwest and east-

southeast.

The terrain maps in Figure 8-2and the gradient profiles in Figure 8-4 and Figure 8-5revealgradual slopes

immediately to each of those directions. The tower is on a small plateau, at least 150 m from any significant

gradient (5%-10%)in the primary wind direction, and 320 m or more from any slope (5%-10%) in the secondary

direction.

Figure 8-4 - Terrain contours, and Google-Earth terrain profile along the primary, west-northwest wind direction

52

Figure 8-5 - Terrain contours, and Google-Earth terrain profile along the secondary, east-southeast wind direction

53

8.3 Shear Characteristics – All Valid Data between 2012-Oct-05 and 2013-Oct-04

The charts in this section characterize the wind shear measured at the site taking into account validated data during the one-year period from 2012-Oct-05 to 2013-Oct-04. Shear exponent values were calculated for all time steps reporting concurrent validated data for the derived variables UT and UB, where the 39 m wind speed reported was at least 3 m/s. Over this time period there were 52,560 time steps, and 35,641 of those met the wind speed threshold. The choice of date range was selected for two main reasons: data recovery rates at the 39 m level were highest for both sensors during this 12-month period (Figure 6-1 and Figure 6-6 ); and quality control could be carried out thoroughly because all sensors were performing properly, including the vane, allowing for a proper assessment of shading, distortion effects, and weather-related sensor underperformance. Time steps with light winds, below 3 m/s at the 39 m level, were excluded from the characteristic shear computations as they often reveal either excessively high or false negative wind shear values that may adversely affect the calculations but aren't relevant to turbine power production. Full shear analysis of the derived UT and UB data sets is carried out in section8.3.1, measured shear charts for the physical sensor data are shown in Appendix11.3.1, and11.3.2. Figure 8-6 is for reference in the discussions below.

Figure 8-6 - U1, U2, UT number of wind events by D1 wind sector when the 39 m wind speed is at least 3 m/s

(up to 9000 events per sector)

54

8.3.1 Shear Calculated Using Derived Variables UT & UB

Figure 8-7 - Characteristic shear charts, where UT > 3m/s α = 0.0813, z0 = 0.00012

Top: wind speed profile to 60 m; shear exponent by sector; Middle: diurnal shear exponent profile; monthly wind speed profiles;

Bottom: monthly shear exponent; UT, UB wind speed roses

Wind speeds and shear were generally higher in the winter months, though low data returns from the physical

sensors affect those numbers. Figure 8-7reports an unexpectedly low roughness length for the site; 0.0001 is

closer to the roughness length for a water body, such as a lake. The characteristic shear exponent is also low,

which is interesting given that shear calculable rates were almost zero during the winter months (Figure 8-8) when

55

the actual surface roughness would have been expected at its lowest due to snow cover. This suggests that

atmospheric stability, which tends to reduce wind shear by transferring horizontal momentum vertically, was an

influencing factor on shear during the summer months. June and July reported the lowest shear values, which is

evidence of the stability influence. Section 9 on turbulence intensity indicates an increase in TI during the summer

months, as expected.

Figure 8-8 - UT (blue) and UB (purple) recovery rates and shear exponent calculable occurrences, by sector and month, where UT > 3 m/s

56

Figure 8-9 - Heat chart of shear exponent by month and hour of day, where UT >3 m/s

The summer stability effect on wind shear is also revealed in the heat plot in Figure 8-11. The lowest wind shear

values, indeed negative values, were reported in the afternoons during the summer months. Calculable shear rates

were low from2012-Dec through 2013-Mar, as well as 2013-Oct, primarily due to the absence of data because of

low returns at the 16.5 m sensor level.

Figure 8-10shows that the shear was fairly consistently below 0.2 in most wind sectors throughout the year, and

again, low returns in 2012-Nov, 2012-Dec and 2103-Mar suggest outlying vertices in the polar plot are probably

not accurate reflections of the monthly shear from a sector.

Figure 8-10 - Shear exponent by month and sector, where UT >3 m/s

57

Figure 8-11 - Shear exponent by month and sector, where UT >3 m/s

Figure 8-11 shows that the wind shear tended to be higher from the southerly sectors when the winds were at

their peak for those directions. In these sectors the terrain is slightly more rugged and may have turbulence-

inducing characteristics. When a met tower is atop a ridge in significant winds, and upwind of that location is a

terrain feature such as a slope, a ridge, or a depression, then if the tower is beyond the mechanical turbulence

zone it could report a higher wind shear profile than it would in flat terrain24

.Figure 8-12and Figure 8-13show the

terrain profiles in the 225° and 135°directions respectively; each has a number of potentially important features,

but it should be noted that all the gradients along the paths are slight and so are the shear extremes shown above.

24

Rohatgi& Nelson 134

58

Figure 8-12 - Terrain profile to the southwest of the met mast, maximum slope is 11% or 6°, and the marker is shown at

1000 m from the tower

Figure 8-13 - Terrain profile to the southwest of the met mast, maximum slope is 18% or 10°, and the marker is shown at

1000 m from the tower

59

Figure 8-14 - Shear exponent by UT, where UT > 3 m/s

Figure 8-15 - Shear exponent frequency distribution, where UT > 3 m/s

60

8.4 Extrapolation to Hub Height

Figure 8-16 - Post-QC wind shear exponent α vs. UT prior to 2013-Dec-12

Colour is scaled by UB

Figure 8-16shows that the bulk of the extreme shear values occurred when wind speeds were low. When extrapolating the measured wind speeds to the 60 m hub height, the wind shear exponent α was restricted to values between -0.05 and 0.4.

Figure 8-17– 60 m Hub height wind and recovery statistics

Left: all valid data used; 2012-Oct-05 to 2012-Oct-2013 data only

61

The data set at the 60 m hub height was referred to as p60 and was generated using Windographer's vertical

extrapolation tool, which was set to use the power law to calculate shear exponents for every time step where

both UT and UB were valid, and to fill in the gaps using other statistics where required.

The important hub height wind speed charts are shown in Figure 8-18, Figure 8-19, and Figure 8-20

Figure 8-18 – Diurnal Hub height (pink) and UT (blue) and UB (purple) wind speeds

Figure 8-19 - Monthly Hub height (pink) and UT (blue) and UB (purple) wind speeds with recovery threshold for averaging

relaxed to 40%

62

Figure 8-20 – Hub height wind speed histogram

Best-fit Weibull parameters: shape k=2.03; scale c = 8.48 m/s

63

Figure 8-21 -

64

9 Turbulence Intensity (TI) Turbulence Intensity (TI) is a dimensionless parameter used to quantify the degree to which the measured wind climate is affected by separated flow or turbulent eddies, which are most commonly caused by flows over rough surface elements or by solar irradiation of the ground creating thermal instability (i.e. turbulent mixing). These types of atmospheric phenomena can put mechanical loads on turbines which will cause wear and tear over time. Turbulence generally reduces wind shear by mixing horizontal momentum vertically. TI is calculated for a 10-minute data time step as the ratio of standard deviation of the wind speed to the mean wind speed recorded.

IEC 61400-1 Edition 3 classifies wind turbulence classes with reference to a mean TI-15 value

25: the expected mean

TI value for wind events falling into the 15 m/s speed bin. The mean TI-15 values for the classifications are: Class S, > 0.16; Class A, 0.14 to 0.16; Class B 0.12 to 0.14; and Class C, 0 to 0.12. The least turbulent class is C. Other IEC documentation makes reference to so-called representative TI-15. The representative TI of a binned period of wind speed data is the mean of the TI values recorded over that period and in that bin plus 1.28 standard deviations. This report quotes both mean and representative TI statistics, though representative TI-15 is a good benchmark to qualify a turbulence class as it is also the 90th percentile of the TI-15. As in section 8 on wind shear, this section of the analysis will only use data collected between 2012-Oct-05 and 2013-Oct-04

25

IEC 61400-12-1, 22

65

9.1 Mean and Representative TI Characteristics

Figure 9-1 – Representative TI and IEC Turbulence Classes A, B and C;

Top: U1, U2 Bottom: U3, U4

Figure 9-2 Mean (green) and Representative (purple) TI by wind direction

66

Figure 9-3- Diurnal Mean (green) and Representative (purple) TI

Figure 9-4 - Mean (green) and representative (purple) TI by month

During the summer months, when this site would have been snow-free, solar heating of the ground would have resulted in a less stable atmosphere around the met mast, with turbulent mixing reducing wind shear, as noted in section 8. During the winter the snow would have reflected much of the sunlight, producing for a less turbulent

67

wind regime. Figure 9-4 and Figure 9-3 show the stability-related turbulence regime reflected in the monthly and the mean diurnal data. After quality control U1, the R.M. Young propeller anemometer typically reported lower TI than any of the NRG cup anemometers. This was true regardless of boom orientation or sensor height. This type of anemometer is less susceptible to the influence off-horizontal flows on its reported 10-minute standard deviation than a cup anemometer. This may be partly because of its axis orientation and blade design, but may also be because it has a longer distance constant than the cup anemometers (2.7 m vs. 2.36 m). The distance constant relates to the length of fluid flow which must pass to influence the response of an anemometer

26. Realistically, the R.M. Young sensor

does not report TI in an equivalent fashion to the NRG cups; the readings are probably not directly comparable.

Figure 9-5 - Bulk TI statistics by sensor

26

http://www.renewablenrgsystems.com/sublayouts/C2ECFSpecificationGlossary.aspx?pid=SPEC-I:5966%20G:7%20S:43

68

9.2 TI at the Hub Height

Figure 9-6 - Hub Height TI (HH-TI) characteristics

Top: Representative HH-TI vs. wind speed with U1-SD (R.M. Young); and with U3-SD Middle Left: HH-TI with U1-SD vs. HH-TI with U3-SD

Middle Right: HH-TI rose with U1-SD (brown), with U3-SD (purple) Bottom: Representative and mean HH-TI roses with U1-SD and with U3-SD

69

Turbulence intensity at the hub height is represented by a surrogate ratio of the 39 m sensor level standard

deviation to the extrapolated hub height wind speed at 60 m. This TI value should uphold the principle that TI and

turbulent mixing generally decreases with height as eddies dissipate as they dissipate in the boundary layer;

further away from the influence of solar heating on the ground, the air should be less turbulent.

For this assessment, hub height TI results have been presented with U1-SD used in the calculations, as well as with

U3-SD replacing U1-SD. Regardless, the TI class of the site is still well within the C category.

Figure 9-7 - Hub height TI statistics with U1-SD (left) and with U3-SD (right)

It is clear from the bottom-left graphic in Figure 9-8 that turbulence intensity changes in unison with the

temperature at the site.

70

Figure 9-8 - Hub Height TI (HH-TI) characteristics Top: Representative and mean diurnal HH-TI with U1-SD (R.M. Young); and with U3-SD

Middle: Representative and mean monthly HH-TI with U1-SD; and with U3-SD Bottom Left: Diurnal TI for U1, U2, U3, U4; and hub height with U1-SD (brown) and with U3-SD (purple)

Bottom Right: Monthly TI for U1, U2, U3, U4; and hub height with U1-SD (brown) and with U3-SD (purple)

71

10 Climatological Adjustment

10.1 The Long-Term Reference Data Set

Environment Canada (EC) operates the Trail Valley climate monitoring station, 24 km southeast of Storm Hills, and

45 km north and east of Inuvik. The station has a heated Vaisala ultrasonic anemometer which would commonly

be mounted 10 m above the ground, though Figure 10-3shows it was difficult to confirm that. Regardless, the

positioning of the station is less than ideal for use in industrial wind resource monitoring as it is fairly close to the

ground and so likely heavily influenced by shear, veering or backing of the wind with respect to the 60 m target

height, and thermal turbulence in the summer.

Figure 10-1 - Storm Hills and Trail Valley referenced to Inuvik, NT

The reference data set was reported in 1hr time steps, and consisted of the following useful variables:

R-U1: the average horizontal wind speed recorded in the two minutes at the end of each observation hour

(i.e. average from hr:58:00 to hr:59:59).

From 2000-Jan-01 the wind speed data was measured in knots rounded, and then converted to km/h and

reported as rounded to the nearest whole-numbered value. However, as of 2013-Dec-12 the station

started both measuring and reporting wind speeds rounded to the nearest km/hr, without converting

from knots. This latter data was ignored as it was outside the time period concurrent with the target data.

Interestingly, 2013-Dec-12 is the same day the R.M. Young sensor was damaged at Storm Hills.

72

R-D1: the average wind direction referenced to true north and recorded in the two minutes at the end of

each observation hour. The data was recorded in 10 degree bins from 10° to 360°. A value of 0° was a flag

for calm winds.

R-T1 : average temperature in °C (likely recorded in the two minutes at the end of each observation hour)

Figure 10-2 - EC Trail Valley quality-controlled reference wind speed frequency histogram, with 1km/hr wind speed binning

Figure 10-3 - The ultrasonic anemometer with the nearby structure in background

73

Figure 10-4 - Satellite imagery of the region surrounding the EC reference data source

In close proximity to the ultrasonic sensor mount is a much larger tower which would certainly cause wind shading

when upwind of the EC sensor. Water bodies in Figure 10-3 and Figure 10-4 suggest it is to the north-northwest of

the EC anemometer.

74

10.2 Comparison of the Concurrent Target and Raw Reference Data Sets

Figure 10-5 - Storm Hills target data hourly wind speed averages by month, with temperature

2012-Oct-04 to 2013-Dec-11

Figure 10-6 - Trail Valley reference data hourly wind speed averages by month, with temperature

2012-Oct-04 to 2013-Dec-11

75

The two sites show similar mid-to-late day speed profiles by month. However, being close to the surface, the Trail

Valley reference data (Figure 10-6)did not show any nocturnal speed-up during the summer (May to August),

rather it slowed down in the evening when the sun was lower in the sky (or set), and there would have been

increased stability and shear in the lower boundary layer.

Figure 10-7 - Target (left) and un-quality controlled reference (right) wind direction frequency roses

2012-Oct-04 to 2013-Dec-11

Figure 10-8 - Storm Hills target data, wind direction frequency roses by month

2012-Oct-04 to 2013-Dec-11

76

Figure 10-9 - Trail Valley target data, wind direction frequency roses by month

2012-Oct-04 to 2013-Dec-11

The monthly average wind frequency roses showed similarities during the winter months (September to April),

with the exception of February. The summer months showed a significant north-northeasterly component at the

EC reference site; the same direction where the sensor would be shaded by the upwind structure. This wind may

be a lake breeze caused by solar radiation-induced temperature gradients between the land and the large water

body to the northwest of the Trail valley site. The Storm Hills site may also see such an effect from the same water

body, but because of its location, the winds would be westerly.

This analysis suggests that data from the summer months may reduce the quality of any overall correlation

between the concurrent target and reference data sets.

77

10.3 Quality Control of the Reference Data

The EC reference data set was collected in a significantly different fashion to the target at Storm Hills. The direction

data was of relatively low precision, and there was no redundant sensor at the site, so only a simple quality control

process was applied to the sonic anemometer.

The anemometer wind speed data was left scaled in km/hr. If the wind direction was reported as zero (calm), then

all sensor data was discarded, and if the sonic anemometer reported zero wind speed, then all sensor data was

discarded.

As well, the proximity of the larger tower warranted the removal of all data reporting winds from sectors centered

around the north-northwest. The range of sectors potentially affected was unclear, but it was decided that all data

winds reported from within 45°of 22.5°true should be removed. This was also convenient given that the summer

months at the reference site were shown in section 10.2 to have components from this direction which were likely

uncorrelated to the measurements at Storm Hills.

Figure 10-2 and Figure 10-10 show the results from the reference data quality control. The conversion from knot

measurements to rounded km/hr values makes the EC reference wind speed histogram appear with bin breaks and

a giant spike; the Weibull fit statistics in the chart should be ignored.

The red circle in the chart at the top-right of Figure 10-10 shows that after quality control there were a significant

number of concurrent data points where the winds at the target site were from northeasterly to southeasterly

directions, while the reference site experienced winds from the west to the northwest. This discrepancy can be

seen as a distinguishing feature between the two sites in the wind rose comparison chart. There were relatively

stronger westerly and northwesterly components in the reference wind rose, while the target wind rose had

stronger components from all easterly directions. Clearly the winds at the time steps circled in red were

uncorrelated and so could have adversely affected the correlation used in the final climatological adjustment.

Comparison of Figure 10-8 and Figure 10-9 suggest that the uncorrelated events may have occurred between

October and April. As well, when the comparison time-step is increased, the effect of the uncorrelated direction

data points is reduced, which is reflected in the table in Figure 10-11 showing that the linear least-squares R2 value

improves with larger steps.

78

Figure 10-10- Quality-controlled 1-hour linear least-squares data set comparison between target and reference data sets

single direction sector bin Overlap period 2012-Oct-04 to 2013-Dec-12

Top: speed and direction target vs. reference scatter plots Middle: target (blue, green) and reference (black, orange) speed frequency profile and frequency roses

Bottom: monthly and sectored wind speeds

79

10.4 Measure-Correlate-Predict (MCP) Process

Figure 10-11 - Preliminary linear least-squares correlation statistics

The data sets differed in their collection and averaging periods: the Storm Hills target data was made up of 10-

minute average values which were then averaged for correlation purposes to 1-hour values at the shortest; while

the reference data were 2-minute averages taken at the end of every hour. Windographer offset the EC reference

data set backwards in time by 40 minutes to maximize correlation potential. This likely accounted for differences

owing to the relative proximity of the sites, as well as to account for the different sampling and averaging

techniques.

Figure 10-11 compares some basic correlation statistics for various comparison time steps where a simple single-

sector linear least-squares correlation regime was used. It was decided that 24 hour time-step was too long a

period, as it would not capture any of the nocturnal wind shear effects noted in the shear analysis. 3-4 hours was a

reasonable time-step in order to smooth out the sampling from in the EC reference data set. The 8 and 12-hour

comparison periods offered small sample sizes and were thought less likely to capture diurnal variations in the final

analysis. The one hour time-step data set was retained for use with a robust matrix time-series correlation

algorithm offered by Windographer.

The bi-polar nature of the wind frequency roses for both the quality-controlled reference and target data sets, and

the generally good R2 values of the direction correlations in Figure 10-11 implied that sector-wise binning of the

data was not necessary. However, the correlation statistics for a linear least-squares correlation regime using 4-

sector binning is offered in Appendix Figure 11-39. Using 4-sector binning, R2 values for the important wind sectors

(135°-315°) were extremely good for winds from the western sector, but not so good from the eastern or southern

sectors, regardless of the correlation time step.

After further analysis, it was decided to work with only one wind direction sector for correlation between the wind

speed data sets. However, for wind direction correlation 12 sectors were used in the 1-hour comparisons, and 4

sectors were used for the 3-hour and 4-hour comparisons.

A number of correlation algorithms were tested against one another for each of the 1, 3, and 4-hr comparison time

steps. Ultimately, in all three cases, a matrix time-series correlation algorithm was employed27

. In each of those

algorithms the target and reference wind speed data sets were binned in 1 m/s and 5 km/hr (1.39 m/s) sets

respectively. The improvement in the shape of the reference data set histogram with this binning scheme can be

seen in Figure 10-12.

27

Windographer 3.2.5 documentation: Matrix Time-series Algorithm

80

Figure 10-12 - EC Trail Valley reference histogram, with 5 km/hr wind speed bins

Figure 10-13 - Storm Hills target histogram, with 1 m/s wind speed bins

81

Figure 10-14 shows the results of correlation algorithm testing for linear least-squares and matrix time-series

techniques. In all, three tests were employed: the first alternated between five hour periods of data correlation of

the target and reference sets, followed by a five hour synthesis of the target data set and analysis of the resulting

error statistics; the second and third tests used ten and fifteen-hour alternating periods in the same fashion.

The 1-hour comparison time-step correlation was found best suited to a matrix time-series algorithm employing an

intermediate-stage analysis technique which binned data using a moving average of values covering three hours

worth of time-steps. The 3-hour and 4-hour comparison steps used the same algorithm, but with an intermediate-

stage that binned data using 9-hour and 12-hour moving averages respectively. The synthesis testing error

statistics quoted in Figure 10-14 are: the mean bias error (MBE); the mean absolute error (MAE); the root mean-

squared error (RMSE); and the error in the frequency distribution (DISE).

Figure 10-14 - MCP testing and error statistics; linear least-squares vs. matrix time-series algorithms

The correlation statistics for the wind direction comparisons can be found in Appendix 11.4. It should be noted

that the MCP process may have been improved if the quality control of the reference data set had included a data

selection step which excluded a time steps when the wind direction was not reasonably well correlated with the

target wind direction. This was an oversight on the part of the analyst and could have been done with a scatter

plot analysis but time did not allow for any modification to the analysis.

82

10.5 Characterizing the Long-Term Wind Climate

The final summary of the three correlation schemes are highlighted in the figures below.

Examination of Figure 10-10 and the summary figures in this section suggest the wind regime during the period of

data collection was not normal from a climatological perspective. Both the target and reference data sets show a

significantly different month-by-month wind speed profile recorded during the overlap period than the reference

set shows over the long-term.

The testing statistics (Figure 10-14) showed little difference in efficacy between the three matrix time-series MCP

schemes employed. The mean of monthly-mean (MoMM) long-term wind speeds at 60 m were estimated to be

7.87 m/s, 7.73 m/s and 7.96 m/s for the 1-hour, 3-hour, and 4-hour comparison time-steps respectively, though a

thorough uncertainty analysis should be carried out at some point because the synthesis error statistics previously

quoted are significant; the by-product of a less-than ideal reference data set. The three-hour time-step MCP

statistics were quoted in the executive summary.

83

Figure 10-15- 1-hour comparison time-step MCP statistics

Figure 10-16 - Summary charts for the 1-hour comparison time-step MCP results

Target data set processed to 1-hour (blue), and the final synthesized target data set (red) Top: diurnal profiles; wind speed frequency profile

Bottom: monthly wind speed profile; wind frequency roses

84

Figure 10-17 - 3-hour comparison time-step MCP statistics

Figure 10-18 - Summary charts for the 3-hour comparison time-step MCP results

Target data set processed to 3-hours (blue), and the final synthesized target data set (red) Top: diurnal profiles; wind speed frequency profile

Bottom: monthly wind speed profile; wind frequency roses

85

Figure 10-19 -- 4-hour comparison time-step MCP statistics

Figure 10-20 - Summary charts for the 4-hour comparison time-step MCP results

Target data set processed to 4-hours (blue), and the final synthesized target data set (red) Top: diurnal profiles; wind speed frequency profile

Bottom: monthly wind speed profile; wind frequency roses

86

10.5.1 3-Hour Time-Step MCP Climatological Adjustment Results

Figure 10-21 - 3-hour comparison time step MCP results

Monthly wind speed histograms

87

Figure 10-22 - 3-hour comparison time step MCP results

Diurnal wind speed profile

The diurnal wind speed profiles in Figure 10-22 reveal that the reference data set effectively washed out the

overnight jet recorded at the Storm Hills target site. A matrix time-series correlation algorithm should preserve

diurnal patterns in the target data set, but this didn't happen for any of the three comparison time-steps studied.

The previously noted quality control issues related to the reference data set reporting un-correlated wind

directions implies that the wind frequency roses in Figure 10-23may present results that are the consequence of

data which should have been removed. Best practice would be to assume the standard wind frequency roses

computed with the Storm Hills data is valid; a fairly good assumption.

88

Figure 10-23 - 3-hour comparison time step MCP results

Monthly wind direction frequency roses

89

11 Appendices

11.1 Data Recovery Rates By Month and Hour of Day

11.1.1 Physical Sensors (data to 2013-Dec-12)

Figure 11-1 - Anemometer U1 Post-QC diurnal data recovery rates (%) by month

Figure 11-2 - Anemometer U2 Post-QC diurnal data recovery rates (%) by month

90

Figure 11-3 - Anemometer U3 Post-QC diurnal data recovery rates (%) by month

Figure 11-4 - Anemometer U4 Post-QC diurnal data recovery rates (%) by month

91

Figure 11-5 - Vane D1 Post-QC diurnal data recovery rates (%), by month

11.1.2 Derived Variables (data to 2013-Dec-12)

Figure 11-6 - Derived variable UT diurnal calculable rates (%) by month

92

Figure 11-7 - Derived variable UB diurnal calculable rates (%) by month

93

11.2 Wind Roses - Validated Sensor Data

11.2.1 Frequency of Wind Occurrence by Sector, and Month (data to 2013-Dec-12)

Figure 11-8

11.2.2 Frequency of Wind Occurrence by Sector, and Hour (data to 2013-Dec-12)

Figure 11-9

94

11.2.3 Frequency of Wind Occurrence by Sector, Month, and Hour (data to 2013-Dec-12)

Figure 11-10 - 2012-Nov

Figure 11-11 – 2012-Dec

95

Figure 11-12 – 2013-Jan

Figure 11-13 – 2013-Feb

96

Figure 11-14 – 2013-Mar

Figure 11-15 - 2013-Apr

97

Figure 11-16 – 2013-May

Figure 11-17 – 2013-Jun

98

Figure 11-18 – 2013-Jul

Figure 11-19 – 2013-Aug

99

Figure 11-20 – 2013-Sep

Figure 11-21 – 2013-Oct

100

Figure 11-22 – 2013-Nov

Figure 11-23 – 2013-Dec

101

Figure 11-24 - Monthly wind frequency roses

102

11.3 Wind Shear: The Physical Sensors

11.3.1 Shear: South-Southwest-Oriented Sensors U1 & U3

Figure 11-25 - Characteristic shear charts, where U1 > 3m/s

α = 0.081, z0 = 0.00011 Top:wind speed profile to 60 m; shear exponent by sector;

Middle: diurnal shear exponent profile; monthly wind speed profiles; Bottom: monthly shear exponent; U1, U3 wind speed roses

103

Figure 11-26- U1 and U3 recovery ratesandshear exponent calculable occurrences, by sector and month, where U1 > 3 m/s

Figure 11-27 - Heat chart of shear exponent by month and hour of day, where U1 >3 m/s

104

Figure 11-28 - Shear exponent by month and sector, where U1 >3 m/s

Figure 11-29 - Shear exponent (colour) by sector and U1 wind speed bin, where U1 >3 m/s

105

11-30 - Shear exponent by U1, where U1 > 3 m/s

11-31 - Shearexponent frequency distribution, where U1 > 3 m/s

106

11.3.2 Shear:North-Northeast-Oriented Sensors U2 & U4

Figure 11-32 - Characteristic shear charts, where U2 > 3m/s α = 0.0755, z0 = 0.00004

Top: wind speed profile to 60 m; shear exponent by sector; Middle: diurnal shear exponent profile; monthly wind speed profiles;

Bottom: monthly shear exponent; U2, U4 wind speed roses

107

Figure 11-33 - U2 and U4 recovery rates and shear exponent calculable occurrences, by sector and month, where U2 > 3 m/s

Figure 11-34 - Heat chart of shear exponent by month and hour of day, where U2 >3 m/s

108

Figure 11-35 - Shear exponent by month and sector, where U2 >3 m/s

Figure 11-36 - Shear exponent by month and sector, where U2 >3 m/s

109

Figure 11-37 - Shear exponent by U2, where U2 > 3 m/s

Figure 11-38 -Shearexponent frequency distribution, where U2 > 3 m/s

110

11.4 Measure-Correlate-Predict Data

Figure 11-39 - 4-sector linear least-squares correlation statistics for various correlation time steps, with both forced and

unforced target intercepts

111

Figure 11-40 - Direction correlation data; 1-hr comparison time-step, 12 sectors

R2 = 0.777

Figure 11-41 - Direction correlation data; 3-hr comparison time-step, 4 sectors

R2 = 0.799

Figure 11-42 - Direction correlation data; 4-hr comparison time-step, 4 sectors

R2 = 0.861